is normalization MUST for integration?

[complexgenome]

Hi satija lab,

I have single cell data from different subjects. I performed preliminary quality control on them. I'd like to integrate them.
The final integrated seurat object needs to have subjects with different cell types and genes with raw counts.

I saw tutorials at:
https://satijalab.org/seurat/articles/integration_introduction.html#introduction-to-scrna-seq-integration-1
https://satijalab.org/seurat/archive/v3.2/immune_alignment.html
https://satijalab.org/seurat/archive/v3.0/integration.html

All these tutorials include step to normalize data before performing the integration.

How can I proceed to achieve the desired integration with final output as raw count matrix object?

thank you.

[timoast]

Yes, you need to normalize the data before running integration. I'm not sure what your goal really is here...if you just need the raw count matrix for each object then there's no need to do integration. If you're looking for some version of a "corrected" count matrix, that's not something that we compute with the Seurat integration algorithm.

[complexgenome]

@timoast thanks for your reply.

I would like to perform deconvolution using single-cell data from multiple subjects. I have single cell data for eight different subjects for a specific tissues region. Per subject one .mtx file

Softwares (music, bisqueRNA, SCDC) for deconvolution want raw counts from different subjects as an input to begin with. They want one file as an input. Whereas I have one object per subject, totaling to eight.

Merging is little sketchy as the count of cells add up as the objects are put together. For instance subj1 has 20K genes and 19K cells, subj2 with 21K genes and 15K cells.

I am unsure what to do and how to proceed at this point.

Thank you again.

[timoast]

I think this is a question to ask the developers of those specific methods (music, etc), about how they handle data from multiple samples.